learn-preference
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill uses a domain-specific language (I-Lang) to set behavioral flags for the agent. These flags, such as observing without asking, applying preferences silently, and saving to global preferences, are consistent with the stated goal of learning user preferences over time and do not exhibit malicious behavior.
- [INDIRECT_PROMPT_INJECTION]: Because the skill is designed to automatically ingest and learn from user interactions, it creates an attack surface where external data (like code being reviewed) could attempt to influence the agent's future behavior. However, the configuration includes a resolution policy where explicit user instructions override learned preferences, which acts as a mitigation.
- Ingestion points: User interactions, code style, and project structure are processed to learn preferences.
- Boundary markers: Not explicitly defined in this configuration file.
- Capability inventory: The skill is limited to updating an internal global preference store.
- Sanitization: Not explicitly defined; relies on the agent's underlying preference management system.
Audit Metadata